|
Huashi Self-Developed 130-200GHz G-Band Broadband Power Amplifier: +15dBm Psat, 15dB Gain, WR-05 WaveguideShenzhen Huashi Digital Technology Co., Ltd. independently develops, manufactures and sells the 130-200 GHz G-band broadband power amplifier module: +15 dBm typical saturated output power (Psat), 15 dB typical small-signal gain, 70 GHz of instantaneous bandwidth, WR-05 waveguide interface, high linearity and low VSWR, intended for millimeter-wave imaging and Foreign Object Debris (FOD) detection. This English technical note turns the published specification into a complete engineering walkthrough: how to read the five decisive numbers on the datasheet, how to split 15 dB of gain across two stages, how to match 70 GHz of bandwidth at wavelengths under 2 mm, and how every indicator is verified on the bench and held across mass production.
Note: typical values are measured at +25 ℃ with nominal bias. Custom frequency ranges, gain and interface options are available for project-based customization. 1. Reading the Spec Sheet: Five Numbers Decide the Architecture70 GHz of bandwidth. Centered at 165 GHz, the fractional bandwidth is 0.424 - firmly ultra-wideband. No single die delivers gain, power, efficiency and stability simultaneously across 70 GHz; the S-parameters necessarily slope, and the design task is deciding where to place the performance and how to flatten the slope with passive networks. The bandwidth also pushes harmonics out of reach: the second harmonic of a 200 GHz carrier lands at 400 GHz, beyond conventional test equipment, so chip-level harmonic content must be characterized first with a spectrum analyzer plus harmonic mixer before deciding how much on-chip harmonic suppression to integrate. +15 dBm is Psat, not P1dB. +15 dBm is 31.6 mW; the +12 dBm floor is 15.8 mW. But Psat is the saturation knee. System integrators care about P1dB, typically 2-3 dB below Psat for millimeter-wave dies. To guarantee module-level Psat of +15 dBm, the final die must deliver +19 to +20 dBm at its own plane, leaving roughly 4 dB for interstage loss, mismatch, temperature drift and lot spread. Referred to a 50-ohm load, +15 dBm is about 1.26 V peak and 25.2 mA peak under perfect match. 9 V at 0.45 A. That is 4.05 W of DC input against 31.6 mW of RF output - a whole-module efficiency of about 0.78%. It looks absurd, but it is the physics of this band: transistor ft approaches its limit, and per-stage PAE of 5-12% is normal. Nearly 4 W becomes heat. The gold plating on the copper housing is not cosmetic - it prevents oxidation that would degrade contact resistance and thermal performance. Die-to-case thermal resistance must be kept in the single-digit ℃/W range. Three voltage levels. +9 V typical, +12 V maximum allowed, +15 V absolute maximum. Ask the customer which supply they will use - a regulated bench supply with low output impedance, or an AC/DC adapter whose soft start can overshoot 2-3 V with a turn-off tail. The rule is absolute: drain voltage, including turn-on transients, must never exceed 12 V. Avalanche damage in the high-field region is cumulative - a 0.5 V overshoot today becomes a field failure three months later. Input-side TVS clamping plus RC soft-start holding the slew below 1 ms/V is mandatory. The 0 to +50 ℃ box. It means industrial-grade screening is sufficient, and it means the gain drift across that box must be absorbed by internal temperature compensation - gate voltage trimmed with temperature to hold gain and power. The +12 dBm RF input limit says two things: the upstream multiplier or upconversion chain must never exceed +12 dBm at the input, and with only 3 dB between input limit and rated Psat, this module is inherently a final-stage PA - do not cascade another stage after it. 2. Link Budget: Splitting 15 dB of Gain Stage by StageA single-stage solution needs one die with 15 dB gain and +18 dBm or more Psat at 200 GHz - such dies exist but are expensive, long-lead and offer no recovery path if a lot shifts. The proven architecture is two stages: a driver die with 9-10 dB gain and +12 to +14 dBm Psat, followed by a power die with 6-7 dB gain and +20 dBm Psat. Summing 15-17 dB of small-signal gain and subtracting 1-1.5 dB of interstage matching and bondwire loss lands the net gain at approximately 15 dB. Why does the final die need +20 dBm when the module is specified at +15 dBm? The 5 dB of margin is consumed deliberately: 1-1.5 dB in interstage networks and bondwires, 0.5-1 dB in the output match including DC blocks and bias injection, 1 dB for hot-case power droop, and 1-1.5 dB for lot spread and VSWR-induced mismatch. This is not conservatism - it is the difference between passing and failing in production. One more subtlety: cascaded stages love to pile up an interstage gain peak at some frequency, saturating early and wrecking flatness. Deliberately lossy interstage matching - giving up 0.5 dB - buys back a flat band. Take the 0.5 dB hit. The TBD noise figure still deserves two numbers in your head: small-signal InP or GaAs dies in this band achieve 4-6 dB NF, and a large-signal final-stage PA sits at 8-12 dB - nothing to agonize over, but the driver-stage NF multiplied by its gain sets the phase-noise floor of the whole transmit chain. 3. Die Selection: Choose by Process, Not by Part NumberThree process routes compete for 130-200 GHz power amplification:
Five hard indicators when screening a candidate die: (1) S21 at 130/140/160/180/200 GHz, small-signal and large-signal, not just a typical value; (2) Psat and P1dB versus frequency - many curves stop at 170 GHz, so demand the 200 GHz number; (3) quiescent point - 3-5 V drain, 150-250 mA/mm of gate periphery is the common domestic window, with 0.4-0.8 W/mm (GaAs) or 0.3-0.6 W/mm (InP) power density; (4) stability factors K greater than 1 and B1 greater than 0 across the whole band, otherwise budget 0.5-1 dB of gain for on-chip resistive stabilization; (5) die size and bond-pad positions - at 200 GHz the bondwires are circuit elements, not wires. Domestic second-sourcing is never a pin-for-pin swap. The working procedure: load the new die's S-parameters in ADS, compare the load-pull optimum impedance region against the incumbent; if the optimum circles differ by more than 20% in radius, redesign input and output matching - otherwise trim compensation elements only. Where vendor datasheets lack temperature and lot-consistency data, buy three lots and measure S-parameters and power at -40, +25 and +85 ℃ yourself. That database is the only reliable verdict on whether a supplier can enter volume production. Huashi maintains qualified domestic die supply channels across all three process routes and performs this verification in-house for every project. 4. LNA and Filters: Underrated, Failure-Deciding BlocksFor transceive architectures the front end adds an LNA. InP achieves 3-4 dB NF but is expensive with unstable lead times; GaAs delivers 4.5-6 dB NF with stable supply. Screen by NF, gain, OIP3 and P1dB - OIP3 minus gain gives IIP3, which sets interference resilience. Any switch or circulator in front of the LNA must stay under 1.5 dB of insertion loss or it degrades system sensitivity directly, and needs more than 30 dB of isolation to protect the receive path from the transmit power. Filters appear in two places: interstage, suppressing harmonics and spurious before the next stage amplifies them, and at the output, meeting system-level out-of-band rejection. Waveguide filters offer low loss and high power handling at the cost of volume; microstrip filters are compact but lossy - a few millimeters of microstrip at 200 GHz can burn 0.5-1 dB. Write the numbers into the filter specification yourself: in-band insertion loss under 1 dB, better than 30 dB rejection at 250-300 GHz to kill the second harmonic, more than 20 dB across 260-400 GHz, and in-band return loss better than 15 dB. Do not let the vendor pick "their best" numbers. 5. Schematic: Four Blocks, Each One Assignable to a PersonRF main chain. Input waveguide transition, input match, driver die, interstage match with equalization, final-stage match, output match, output waveguide transition - with insertion-loss budgets annotated right on the schematic: 0.4 dB input transition, 0.6 dB interstage, 0.5 dB output match, 0.4 dB output transition, 1.9 dB total passive loss. When any stage exceeds budget, one glance at the schematic locates the problem. Bias circuits. 9 V enters through ferrites and decoupling capacitors, then splits to both stages. Each stage needs a gate-bias network - the gate of entry for millimeter-wave dies. Gate voltage is negative, typically -0.3 to -0.8 V, and must come from a DAC or adjustable reference, never a fixed divider, because it doubles as the temperature-compensation and power-trim handle. Series resistors of tens of kilo-ohms with a 10-100 pF bypass at the gate pad form an RC low-pass with corner below 100 MHz, filtering low-frequency oscillation and noise on the bias line. Power and protection. Sequencing is life-or-death: gate voltage must establish before drain voltage, and drain voltage must be removed before gate at shutdown - violate it once and the die usually dies. An enable-controlled timing circuit delays drain by 1-2 ms after gate. Protection is three items: TVS clamping (breakdown above 12 V, clamping below 15 V), MOSFET soft-start stretching rise time to 1-5 ms, and overcurrent detection - sense resistor plus comparator at 0.6 A threshold with immediate shutdown. Control and monitoring. Exposed lines: enable, gate control (analog or SPI), temperature sampling, current sampling (low-side sense into ADC), and power detection through a 20-30 dB directional coupler after the final stage, rectified into the ADC for closed-loop output power control. A 0.5 ℃-accuracy digital sensor on the housing is sufficient. 6. PCB and Packaging: Drawing Transmission Lines at 130 GHzSubstrate. Quartz (permittivity 3.78, loss tangent around 0.0001) buys 0.5-1 dB of loss margin for long, complex matching networks; alumina (permittivity 9.8) packs lines shorter and tighter where board area is scarce, but its per-millimeter loss at 200 GHz is visibly higher and must be charged to the link budget. 50 ohms in numbers. On 0.127 mm (5 mil) quartz with half-ounce copper, a 50-ohm microstrip is about 0.38 mm wide with effective permittivity 2.6 - guided wavelength near 0.58 mm, quarter-wave just 0.145 mm at 200 GHz. Below 150 micrometers. Quarter-wave compensation sections are the same physical scale as bondwires, so lumped elements and short stubs replace textbook transformer sections. Ground. RF ground at 200 GHz is not a copper pour - it is dense via fences. Stitch vias along both sides of every RF trace at 100-150 micrometer pitch to block parallel-plate and higher-order modes; spacing that is too loose shows up as inexplicable gain ripple and oscillation. Die ground pads get full via arrays at 150 micrometers or less, tied through every layer. Heat path. The die sits on a molybdenum-copper or copper-tungsten carrier (CTE matched to the die substrate, or thermal cycling will crack it), soldered or bonded to the copper housing. Target 8-12 ℃/W die-to-case: 4 W raises junction 32-48 ℃ above case, so a 50 ℃ case means 82-98 ℃ junction - within the 150 ℃ GaAs limit but not comfortably. A heatsink is mandatory. Waveguide transition. WR-05, TE10 mode, cutoff near 118 GHz - just below the 130 GHz band edge, which is exactly why it covers the band. E-plane probes (0.3-0.5 dB typical) are preferred over finlines for simplicity and loss. Probe depth, angle and microstrip open-stub length set the match; simulate with metallization thickness and waveguide wall roughness included, or the measurement will drift from prediction. Fab notes. Specify copper thickness and tolerance, dielectric tolerance at plus/minus 5% (plus/minus 10% drifts 50-ohm impedance by 3-5 ohm and detunes the match), 0.5-1 micrometer gold finish, minimum hole copper, minimum line/space capability, and surface flatness. 7. Broadband Matching: The Heart of This DesignKnow the limit first - Bode-Fano. For a given load, the wider the band and the deeper the return loss, the more network complexity is demanded. At 130-200 GHz targeting better than 10 dB return loss across the band, single-section quarter-wave transformers are hopeless; multi-section Chebyshev or maximally-flat structures are required. Move one: multi-section transformers. Cascade sections of differing characteristic impedance, each about an eighth of a wavelength so reflections partially cancel across the band - three sections minimum, sweeping section count and impedances in ADS until S11 stays below -12 dB from 130 to 200 GHz. At 200 GHz a quarter-wave is 145 micrometers, so each section shrinks to 60-80 micrometers and line-width tolerance must hold at plus/minus 5 micrometers. Move two: tapers. Exponential or Klopfenstein tapers give extreme bandwidth and low return loss when space allows - a useful taper is 1-2 mm long at this wavelength. Best reserved for the input and output transitions where room exists. Move three: open stubs. Shunt open stubs are the most flexible tool. A stub slightly shorter than a quarter-wave presents capacitive reactance across the band, canceling the inductive input of a typical die; multiple stubs widen coverage. At these wavelengths stubs are tens of micrometers long - leave manufacturing margin in the layout rather than designing to the very edge of fab capability. Move four: resistive stabilization. Where reactive matching alone cannot flatten a stubborn peak, a few ohms of series resistance or a shunt resistor swallows it at the price of half a decibel. An intentional trade - less gain, guaranteed stability. Load-pull is the foundation. Sweep a grid of complex impedances at the output plane, record power and efficiency at each point, and map the constant-power and constant-efficiency circles - their overlap is the optimum region. Mechanical tuners struggle at 200 GHz; semiconductor-based tuners or on-wafer load-pull are the current practice. Measure 5-9 frequency points with 15-25 impedance points each. The result tells you how matchable the die is: a smooth locus of circle centers with modest radii means broadband-friendly; jumping centers or huge radii means an equalizer or a channelized approach is coming. Input versus output. The output match chases maximum power transfer to the optimum-power circle; the input match chases conjugate match for minimum reflection and small-signal gain - two different impedances, two different networks. The input network also dominates stability; many inexplicable oscillations trace to an input impedance gone bad at some frequency. The interstage network carries three jobs at once: impedance transformation, DC blocking (a 100 pF or larger capacitor in the smallest available package), and stability via small series resistors or series-R/shunt-RC - plus gain equalization to pull down the driver's low-frequency gain hump. Bondwires are the circuit. A single 25-micrometer gold wire, 300 micrometers long, is 0.25-0.35 nH - over 300 ohm of reactance at 200 GHz. Model every bondwire in EM simulation as a distributed element; use two parallel wires or ribbon to halve inductance and spread; freeze arc height, span and wire count in the process documentation so operators cannot improvise. Harmonic trapping by matching. Shape the output network to match the fundamental while deliberately mismatching the second harmonic: a stub that is a quarter-wave at 200 GHz is an open circuit to the fundamental and a half-wave short to ground at 400 GHz, pulling the second harmonic down 5-10 dB before any filter. 8. How Each Specification Is Actually AchievedGain. Calibrate the VNA to the module ports, drive at -30 to -20 dBm confirmed small-signal, sweep the whole band. If gain is short: check DC operating points first, then per-stage S-parameters, then match-network loss and bondwires. The most common culprit is gate bias not seated where the design assumed. Psat. Power-sweep with a band-appropriate millimeter-wave power sensor, recording input, output and DC current; the knee of the output curve is Psat. Uniformly low across the band points to bias or die-lot issues; locally low points to match-network frequency response. Track PAE alongside - (RFout - RFin) / DCin - and accept that around 1% module-level PAE is simply what this band gives. Flatness. Defined as gain spread across the band, typically 1.5-2.5 dB allowed. A monotonic tilt is die gain slope and calls for a gain equalizer - a resonant absorption network placed interstage, tuned in ADS to hold flatness without adding VSWR. A narrow spike means an interstage resonance or oscillation, fixed with resistive loss at that frequency. Periodic ripple means via pitch is too wide - parallel-plate modes - tighten the fence. Linearity. Two-tone tests with 10-100 MHz spacing give OIP3, normally 6-10 dB below Psat for a final-stage PA and rarely the binding constraint. For communication links ACPR is the hard number: back off 3-6 dB from compression, or apply DPD - though DPD at 200 GHz is extremely difficult in practice, so back-off remains the tool. VSWR and stability. Return loss better than 10 dB at both ports. If it oscillates: abnormal DC current, unexcited spectral lines, input-power-dependent gain. First add isolation and check bias decoupling - do not start by retuning the match; oscillation usually lives outside the design band where the match was never intended to operate. Harmonics. Spectrum analyzer plus harmonic mixer to beyond three times the carrier; require the second harmonic at least 30 dB down and the third at least 40 dB down, achieved by harmonic traps, output low-pass and waveguide filtering in combination. Noise figure. Y-factor method with a calibrated noise source where a receive path exists; below 6 dB NF generally forces InP in the driver position. Full-temperature test. GaAs gain falls 0.01-0.02 dB per ℃; from 25 to 50 ℃ expect 0.25-0.5 dB of gain droop and 0.3-0.6 dB of power droop, recovered by trimming gate voltage tens of millivolts positive with temperature. Calibrate 5-9 temperature points into a lookup table (thermistor-plus-opamp, or ADC-MCU-DAC) and interpolate; a few millivolts per degree, determined by measurement. 9. Demo Board and Prototype: The First Gate from Simulation to MeasurementBefore the formal layout, build a demo board to verify three things: the die's real S-parameters and power curves in this band, the actual bondwire impact, and whether the matching direction is right. Oversize match-element pads for tuning, add test pads, bring out all power and control lines, and keep the board simple - get the main chain working first. On-wafer measurement above 130 GHz uses waveguide-interface probes calibrated to the probe tips with calibration substrates (TRL). Probe position and contact pressure influence the readings, so accept a die only after three repeat measurements agree within 0.5 dB. Bring-up, step by step. First power the supplies unloaded and verify each rail and its ripple (under 20 mV p-p). Second, connect the module, apply gate voltage first, confirm its value, then drain - watching turn-on waveforms for overshoot on a scope. Third, verify quiescent current: 0.45 A typical, and any deviation beyond 20% stops the test. High quiescent current means gate bias or oscillation; low current means an open bond or bias path. Either way, power down and diagnose before applying RF. Then small-signal: sweep S21 and S11 at -30 dBm and compare curve shape with simulation. Then large-signal: power sweep with input, output and current recorded together - a healthy amplifier shows flat gain, gradual compression and current that settles after saturation. A current jump at saturation means oscillation or supply instability. Finally the chamber: 0 to 50 ℃ in 10-degree steps, recording gain, Psat, current and gate voltage - both for compliance and for the compensation table. The classic trap is heat-induced oscillation: stable at room temperature, oscillating hot, fixed with input-side resistive stabilization. The standing troubleshooting order: low output power - drain/gate bias, die lot, match loss, bondwires, transitions; uneven gain - interstage resonance, missing equalizer, die slope; oscillation - bias decoupling, via density, input match, supply ripple; poor VSWR - match networks, waveguide transition, connectors; abnormal current - gate bias, bonds, bias circuits, the die itself; excess harmonics - traps and output filtering; large drift - compensation network and die temperature coefficients. 10. Board Fabrication, Structures and AssemblyWrite the complete fabrication specification before releasing the board: laminate type and thickness, copper weight, finish, hole copper (15 micrometers minimum), minimum line/space (plus/minus 5 micrometers at this band), layer registration (plus/minus 25 micrometers), surface roughness RMS under 0.4 micrometers, warp under 0.5%, and controlled impedance - all in the purchase contract, not left to the vendor's discretion. When boards return, test them before assembly: transmission-line loss, via integrity, match-network S-parameters against simulation. Large deviation means the process parameters fed into the model were wrong - remodel. The copper housing demands WR-05 aperture tolerance of plus/minus 0.01 mm (the narrow dimension is about 0.5 mm - sloppy tolerance shows immediately as VSWR), internal surface roughness Ra under 0.8 micrometers, 0.5-1 micrometer gold plating (too thin oxidizes; too thick wears threads), and a precision-ground thermal mounting face. Die attach for 4 W of dissipation is eutectic or gold-tin solder, not silver epoxy - conductive adhesive is easier but runs 2-3 times worse thermal resistance. Bonding uses 25-micrometer gold wire with 80-120 micrometer arc height, minimum length, multiple wires in parallel; bond parameters (power, force, time) are frozen during process qualification into the work instructions. Every unit gets visual and X-ray inspection: bond count, position, arc height, and die-attach voids under 10%. 11. Test Methods and InstrumentsVNA with the appropriate millimeter-wave extensions, TRL or SOLT calibration matched to WR-05, keeping source power low enough to avoid receiver compression and module self-excitation. Power meter with a band-covering mmWave sensor - thermistor types for accuracy, diode types for swept power (never above their +20 dBm ceiling); waveguide connection repeatability matters, so measure three times and re-measure if readings spread beyond 0.3 dB. Spectrum analyzer with harmonic mixer for harmonics and spurious (reference level high enough to keep the fundamental un-compressed, RBW/VBW narrow at 1-10 kHz), and two-tone setups for IP3. Signal source as a low-frequency synthesizer driving a multiplier chain or a native mmWave source covering -40 to 0 dBm with 0.1 dB steps and power-sweep capability. Oscilloscope at 200 MHz bandwidth minimum with differential probes for turn-on overshoot, ripple, and pulse modulation fidelity. Chamber and burn-in - programmable chamber spanning the working range, plus 48-168 h of powered high-temperature burn-in to screen infant mortality. Automated test is not optional at volume. A controller drives VNA, source, power meter and analyzer through a waveguide switch matrix, measuring every parameter in one fixturing, judging pass/fail and issuing reports automatically - cutting per-unit test time from about 20 minutes to 5. 12. Applications: Where These Numbers Go to Work
13. Seven Field Cases: Everything Above, Connected
14. Mass Production: From One Prototype to One Hundred UnitsDV (design validation) proves the design meets every specification: full-band S-parameters, Psat and efficiency, flatness, VSWR, harmonics, full temperature, vibration, shock and thermal cycling across 5-10 units. PV (process validation) proves the process can repeat: 30-50 consecutive units, statistics on every parameter, Cpk above 1.33, and complete process documentation - assembly SOPs, bond-parameter tables, test specifications, acceptance criteria, traceability records. Incoming inspection is the first gate: DC parameters and sampled S-parameters/Psat on dies, permittivity and loss tangent on laminate, dimensions and roughness on waveguide parts, value and ESR on passives. Critical devices get a 5-10 piece per-lot sample test with a lot archive. SPC closes the loop: X-bar R charts on Psat, gain and current with 5 units per lot - two real catches from trend-watching alone were a drifting gate-bias reference and a laminate thickness shift, both before any specification was violated. Production spread always reduces to three roots, with three matching remedies: die-lot variation - incoming binning with per-bin compensation through DAC-controlled gate trim; assembly variation - parameters frozen in SOPs and enforced by automatic equipment; process variation - tolerances tightened contractually with suppliers (line width, dielectric thickness, hole copper). Applied together they take yield from 60% to 95% and above; the remainder is recovered at final trim - gate micro-adjust or a small match-compensation shim on borderline units. 15. Ten Takeaways for New Engineers
16. Appendix Digest: Components, Layout and Bench PracticeReading a die datasheet. Page one: process, band, typical gain, Psat, DC parameters, package. Then S-parameter curves (watch S21 slope and S11 below -10 dB everywhere), power-versus-frequency (the 200 GHz number specifically), temperature curves across -40 to +85 ℃, and reliability data (HTRB, temperature cycling, mechanical shock). Typical values are measured under the vendor's conditions, not yours - always re-run the S2P file in ADS. Domestic die landscape. InP: the CETC 13th and 55th Institutes lead, with fast-moving private fabless teams behind them. GaAs pHEMT: multiple foundries and fabless suppliers - cheap and plentiful, gain-hungry at 200 GHz. GaN: mature below 40 GHz and in narrowband power, not here. For control silicon: domestic FPGAs (Pango Logos, Anlogic Elf) handle gate-DAC control, temperature/current acquisition and host communication with 5k-20k LUTs; MCUs (GD32, HC32, N32 families) handle system management. The power tree uses domestic bucks and LDOs (SGMicro SGM6012/SGM2036, Silergy SY8088, NanCore SC8101), an SGM3204 charge pump for the negative gate rail, 12-bit multi-channel DACs for gate trim and SAR ADCs for monitoring. Passives: NP0/C0G capacitors only for RF bypass (X7R bias-dependent capacitance destabilizes decoupling), thin-film resistors for bias networks, and transmission-line chokes rather than lumped inductors whose self-resonance falls far below 200 GHz. Layout rules that carry the band. RF layer adjacent to an unbroken ground plane; constant line width with rounded corners; three line-widths of spacing between RF traces; via fences at 100-150 micrometers on both sides of every RF trace and full via arrays under die pads; wide, short power feeds with three-stage decoupling placed in parallel at each stage; low-speed SPI/I2C traces short, 2 mm away from RF, clock lines fence-guarded, 22-33 ohm series resistors to soften edges. Bench practice. TRL calibration to the probe tips; three-measurement repeatability rules (0.5 dB on-wafer, 0.3 dB on power); harmonic mixing for 400 GHz measurements; two-tone IP3 with 10-100 MHz spacing; Y-factor NF; temperature sweeps settling 15-20 minutes per point with the sensor on the housing near the die - recording case temperature, not chamber setpoint. Failure statistics worth memorizing. Bondwire opens and cold joints about 40% of field failures, die-lot issues about 25%, assembly stress cracking about 15%, electrical overstress about 10%, others 10%. Put improvement resources where the distribution points. 17. Supply and CustomizationThe 130-200 GHz G-band broadband power amplifier module is in volume supply. Shenzhen Huashi Digital Technology Co., Ltd. maintains end-to-end independent research, development, manufacturing and sales capability for RF, microwave and satellite-communication amplifier products - from die selection and qualification, broadband matching and EM simulation, thermal and housing design, through automated production test and lot-level consistency control. Custom frequency ranges, gain profiles, output power levels, interfaces and full-temperature specifications are supported on a project basis, with English documentation and data packages available for international customers. For selection and quotation, please contact: Sales Department 18688706870. Website: www.szwisi.com. |